When the problem is present, the total-power spectrum appears to be split into eight bands. By comparing the passband spectrum with a "good" spectrum, it seems like the bands are coming out in the correct order, but with strange baselines.When present, it affects all eight banks identically.
We observed for around twenty hours Fri/Sat/Sun 18/19/20th July, and took literally hundreds of scans, apparently without any problems (I have not review every single scan). We observed on Thursday July 24th, and the problem was present.
What modes does it affect?
It is present in modes 3,4 and 10. We have not yet tested any other modes.